# google-deepmind/synthid-text

Repository: https://github.com/google-deepmind/synthid-text
Canonical: https://ross.abutalabs.com/products/synthid-text
Language: Python
License: Apache-2.0
License Family: permissive
Last push: 2026-07-17T21:40:04+00:00

## Health v2 (maintenance only)
Score: 65/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 93, release rhythm 40, longevity 48
- inputs: {"age_days": 679, "days_push": 47, "days_rel": 657, "gap_med": 0, "n_releases_24m": 4}
- flags: prerelease_only
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1066, forks 99 (observed 2026-08-28T04:03:27.050242+00:00)

## What it is
A reference Python implementation of Google DeepMind's SynthID Text watermarking and detection technology for identifying AI-generated text. It extends Hugging Face Transformers models like Gemma and GPT-2 with watermarking mixins and provides both weighted mean and Bayesian detectors.

## Use cases
- watermark text generated by LLMs like Gemma or GPT-2
- detect whether text was generated by an AI model
- reproduce results from the SynthID Text research paper
- experiment with LLM watermarking techniques in PyTorch
- train a Bayesian detector for watermark detection
- integrate watermarking into Hugging Face Transformers pipelines

## When to choose
- you need a reference implementation for research or reproducibility of the SynthID Text paper
- you want to experiment with watermarking and detection on Gemma or GPT-2 models
- you are exploring AI-generated text detection techniques in Python

## When to avoid
- you need a production-ready watermarking solution (use the official SynthID Text implementation in Hugging Face Transformers instead)
- you require cryptographic security guarantees for the hashing functions used
- you are not working with PyTorch or Hugging Face Transformers

## Facets
- artifact type: library
- maturity: experimental
- function: nlp, machine-learning, security, sdk
- domain: large-language-models, artificial-intelligence, security
- platform: python
- tags: watermarking, ai-generated-text-detection, synthid, huggingface-transformers, pytorch, llm-watermark, text-detection, deepmind, natural-language-processing, research, gpu

## Member repositories
- google-deepmind/synthid-text (main) score 65

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:03:27.050242+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T06:55:24.248657+00:00, confidence not recorded.
  - readme: https://github.com/google-deepmind/synthid-text (fetched 2026-08-28T04:03:27.050242+00:00, sha 78a12861f1ab)
  - registry_pypi: https://pypi.org/pypi/synthid-text/json (fetched 2026-08-29T12:57:23.714900+00:00, sha 644429156885)
- Data as of 2026-08-30T08:39:29.467469+00:00.